[
https://issues.apache.org/jira/browse/PHOENIX-1292?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14147936#comment-14147936
]
Lars Hofhansl commented on PHOENIX-1292:
----------------------------------------
Profiled a bit.
There are two reasons:
# SingleKeyValueComparisonFilter returns ReturnCode.NEXT_ROW from
filterKeyValue causing a seek
# When an empty result for a row is returned in the HBase RegionScannerImpl, it
has to recheck whether it needs to proceed with the next row
Just fixing #1 didn't yield much improvement (and it would be slower when there
are many column, we should pass the SEEK_TO_COLUMN hint around, maybe). #2 is
not easy to fix in HBase.
> Some queries that filter all rows are slow.
> -------------------------------------------
>
> Key: PHOENIX-1292
> URL: https://issues.apache.org/jira/browse/PHOENIX-1292
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
>
> Sorry for the vague title.
> As mentioned PHOENIX-1281, I have seen some runtime anomalies for queries
> that should take similar amounts of time.
> 40m row tables, 4 integer column + integer key. All columns (except keys set
> to 1)
> |select distinct(v1) from x where v1 = 1 |21s|
> |select distinct(v1) from x where v1 <> 1 |50s|
> |select count(\*) from x where v1 <> 1 |45s|
> |select count(\*) from x where v1 = 1 |9.5s|
> Interestingly the queries that filter all rows are 5x slower.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
